ebook img

Anisotropic scaling and generalized conformal invariance at Lifshitz points PDF

4 Pages·0.84 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Anisotropic scaling and generalized conformal invariance at Lifshitz points

Anisotropicscalingandgeneralized conformalinvarianceat Lifshitz 2 0 points 0 2 MalteHenkela,1 andMichel Pleimlinga,b n aLaboratoire de Physique des Mat´eriaux2, Universit´e HenriPoincar´e Nancy I, B.P. 239, a J F–54506 Vandœuvre–l`es–Nancy Cedex, France bInstitut fu¨rTheoretische Physik 1, Universit¨atErlangen-Nu¨rnberg, D–91058 Erlangen, Germany 9 2 3 v Abstract 4 5 AnewvariantoftheWolffclusteralgorithmisproposedforsimulatingsystemswithcompetinginteractions.This 4 method is used in a high-precision study of the Lifshitz point of the 3D ANNNI model. At the Lifshitz point, 8 several critical exponentsare found and theanisotropic scaling of thecorrelators is verified.The functional form 0 of thetwo-point correlators is shown tobeconsistent with thepredictions of generalized conformal invariance. 1 0 / Keywords: Conformalinvariance, Lifshitzpoint, ANNNImodel, correlationfunction, Wolff algorithm t a PACS: 05.70.Jk, 64.60.Fr, 02.70.Lq, 11.25.Hf m - d n Competing interactions are encountered in a previous Monte Carlo studies of the transition(s) o large variety of physical systems such as, among fromthedisorderedhigh-temperaturephasetothe c others,magnets,alloysor ferroelectrics.These in- ordered low-temperature phases exclusively used : v teractionsmayleadtorichphasediagramswitha single-spin flip algorithms [4,5]. Due to the criti- i multitudeofphasesaswellastospecialmulticrit- calslowing-downinherenttothesemethods,those X ical points called Lifshitz points (LP). At an LP, numericalstudieswerelimitedtorathersmallsys- r a a disordered, a uniformly ordered and a periodi- tem sizes. However, large systems are needed for cally ordered phase become indistinguishable [1]. theprecisecomputationofcriticalquantities,such A large number of systems (magnets, ferroelec- ascriticalexponentsorcriticalcorrelators. tric liquid crystals, uniaxial ferroelectrics, block In this contribution, we present a new variant copolymers)havebeen showntopossesanLP. of the Wolff cluster algorithm [6] specifically de- This kind of systems may be mimicked by spin signedforthe simulationofsystems withcompet- modelswithcompetinginteractions.Thesimplest ing interactions near criticality. Similarly, we also of these models is the well-known ANNNI (ax- generalizetherecentlyproposedmethodofEvertz ial next nearest neighbour Ising) model [2,3]. Be- and von der Linden for the computation of corre- causeofthepresenceofthecompetinginteractions, lation functions of infinite systems [7] to systems with competing interactions. As an example, we 1 Correspondingauthor: [email protected] present results obtained at the LP of the three- 2 Laboratoireassoci´e auCNRSUMR7556 Preprintsubmitted toComputer Physics Communications 1February2008 + − − + + + − − − + dimensional ANNNI model, and we shall concen- (a) − + − + − − + + + − trateontheLPcriticalexponentsandcriticalcor- relators. In particular, our data for the scaling of + + + − + − − + + + the two-point correlation functions are consistent + − − + + − + − − − withthetheoreticalpredictionsofgeneralizedcon- − − − − + + + − − + l formalinvariance[8,9]. a − + − + + + − − − + i Problemscomingfromcriticalslowing-downen- x + + + − + + + − + + a countered when using local Monte-Carlo dynam- icsarealleviatedbyusingnon-localmethods,such − − − + − + − − − + as the Wolff cluster algorithm [6]. For the Ising + + + + + − + + + + model with only a nearest-neighbour coupling J, + + − + + + − − + − this algorithm may be described as follows: one choosesrandomlyalatticesite,theseed,andthen builds up iteratively a cluster. If i is a cluster site + − − + + + − − − + (withspins )asitej neighbouringtheclustersite i (b) − + − + − − + + + − i will be included into the cluster with probabil- ity p = 1(1+sign(s s ))(1−exp[−2J/(k T)]). + + + − + − − + + + 2 i j B One ends up with a cluster ofspins having allthe + − − + + − + − − − same sign, see Fig. 1a, and which is flipped as a − − − − + + + − − + l whole.Thiskindofsame-signclustersisobviously a − + − + + + − − − + i not adapted to our problem because of the com- x + + + − + + + − + + a peting interactions. The modified cluster algorithm we propose for − − − + − + − − − + simulating spin models with competing interac- + + + + + − + + + + tions is in the following presented for the ANNNI + + − + + + − − + − model.Ageneralizationtoothermodelsisstraight- forward.TheHamiltonianofthe3DANNNImodel Fig. 1. Typical clusters (gray spins) obtained (a) by the is,withs =±1, Wolff algorithm and (b) by our modified algorithm for xyz systems with competing interactions, here shown in two dimensions for simplicity. The competition takes place in H=−JXsxyz(cid:0)s(x+1)yz+sx(y+1)z+sxy(z+1)(cid:1) theaxialdirection.Thespinenclosedbyacircleistheseed. xyz whole,containsspinsofbothsigns,asshowninFig. +κJXsxyzsxy(z+2) (1) 1b.AtvariancewiththetraditionalWolffmethod, xyz the flipped spins are not necessarily connected by whereas J > 0 and κ > 0 are coupling constants. nearest-neighbour couplings. Ergodicity and de- Intheaxialz-directioncompetitionbetweenferro- tailedbalanceareprovenasusual.Thisalgorithm magneticnearest-neighbourandantiferromagnetic works extremely well in the paramagnetic phase, next-nearest-neighbour couplings occurs, leading in the ferromagnetic phase and in the vicinity of to arichphase diagram[10]. the LP.It has not yet been subjected to stringent Our proposed modified cluster algorithmstarts tests in the modulated regionwhere largefree en- with a randomly chosen seed and builds up iter- ergy barriers make simulations very difficult (we atively a cluster. Consider a newly added cluster pointoutrecentprogressachievedinmicrocanoni- latticesiteiwithspins .Alatticesitej withspin calsimulationsoffiniteANNNIsystems[11]).Note i s nearest neighbour to i is included with proba- thatouralgorithmdiffersfromthesame-signclus- j bilityp =p,whereasanaxialnext-nearestneigh- ter algorithm proposed by Luijten and Blo¨te [12] n boursitekwithspins isincludedwithprobability for the simulation of systems with long-range in- k p = 1(1−sign(s s ))(1−exp[−2Jκ/(k T)]). teractions. a 2 i k B Thus the final cluster, which will be flipped as a Inaninteresting work,Evertzandvonder Lin- 2 den [7] have proposed a new numerical scheme, Table1 EstimatedlocationoftheLPinthe3DANNNImodelde- basedontheWolffclusteralgorithm,forthedirect finedonacubiclattice,asobtainedfromhigh-temperature computation of two-point functions of an infinite expansions (HT)and Monte Carlosimulations (MC). lattice. In their approach, each cluster is started at the same lattice site and not from a randomly κL TL chosensite.Afteracertainnumberofclusterflips, HT[13] 0.270 0.005 3.73 0.03 ± ± the two-pointfunctions canbe measuredwithin a MC [5] 0.265 3.77 0.02 certain region, this region getting larger with in- ± presentwork0.270 0.0043.7475 0.005 creasingnumberofflips.Withthismethod,corre- ± ± lationfunctionsofthe infinitelattice areobtained Table2 attemperaturesabovethecriticaltemperature[7]. Criticalexponents atthe LPof the3D ANNNImodel, as At the critical temperature, finite-size effects are obtained from Monte Carlo simulations (MC) and renor- greatly reduced as compared to more traditional malized field theory (FT). The numbers in brackets give approaches. We have generalized the method of theestimated errorinthe lastdigit(s). [7] to systems with competing interactions by re- α β γ (2 α)/γ β/γ placingthe usualWolffclusterswithourmodified − clusters.Thisenablesustocomputewithunprece- MC [5] 0.19(2) 1.40(6) 0.14(2) dentedprecisiontwo-pointcorrelatorsinspinsys- FT [14] 1.27 0.134 tems withcompeting interactions. FT [17] 0.160 0.220 1.399 1.315 0.157 In the following, we report results of a large- scale Monte Carlo study of the uniaxial LP en- presentwork0.18(2)0.238(5)1.36(3) 1.34(5) 0.175(8) counteredinthe3DANNNImodelwherethenew algorithms introduced above have been used with in good approximationequal to 1/2 [14]. In order greatsuccess[9].Thewholestudytooktheequiva- to take into account the special finite-size effects lentofmorethantwoCPU-yearsonaDECalpha comingfromtheanisotropicscalingattheLP[15], workstation. For the investigation of LP proper- largesystemsofanisotropicshapewithL×L×N ties a precise location of this point is mandatory. spins with 20 ≤ L ≤ 240 and 10 ≤ N ≤ 100 have Onepossibilitytolocatethispointisgivenbythe beensimulated. structure factor S(q) = PhS0Sri exp(iqz) as the Our estimates for the LP exponents α,β,γ are r givenin Table 2. For the first time, these ANNNI transition between uniformly orderedand period- model exponents are computed independently. ically ordered phases shows up as a shift of the They were found by investigating effective expo- maximum of S(q) from q = 0 to a non-zero value nents which yield the critical exponents in the [4]. Having found a reliable value for κ , the LP L limit T −→ T [9,16] (provided that finite-size critical temperature is then obtained by standard L effects canbe neglected). Ourerrorbars takeinto methods. In Table 1, we compare the location of account both the sample averaging and the un- the LP as obtained from our data with previous certaintyinthe locationofthe LP.The reliability estimations.NotethatKaskiandSelke[5]didnot of our data is illustrated by the agreement of the attempt an independent determination of κ but L independently estimated exponents α, β and γ merely determined T for a κ in close vicinity to L with the scaling relation α + 2β + γ = 2 up to κ as obtained from a high-temperature (HT) se- L ≈0.8%.Furthermore,theagreementwitharecent ries expansion [13]. From Table 1 the increase in two-loopcalculation[17]isremarkable. precisionis evident. Fig.2summarizesourdataobtainedfortheLP The uniaxial LP is a strong anisotropic critical point where the correlation lengths parallel and spin-spin correlatorC(r⊥,rk) = hSr⊥,rkS0⊥,0i. It showsthescalingfunctionΦ(u)relatedtothecor- perpendicular to the axial direction diverge with relatorby different critical exponents: ξk,⊥ ∼ |T −TL|−νk,⊥. Thevalueoftheanisotropyexponentθ =νk/ν⊥is C(r ,r )=r−2x/θΦ(rθ/r ) (2) ⊥ k k ⊥ k 3 usedintheinsetofFig.2.Theniceagreementbe- tween theory and numerical data provides strong 1 1 u) evidence for the applicability of the generalized Φ( conformal invariance to the strongly anisotropic 0.5 criticalbehaviourrealizedatLPs.Asimilaragree- ) u ment between numerics and theory has been ob- ( Φ tainedfor the energy-energycorrelationfunction. 0.5 0 0 1 2 3 4 5 6 7 Inconclusion,wehavepresentedanewvariantof 1/u1/2 theWolffclusteralgorithmdesignedforthe study ofsystemswithcompetinginteractions.Appliedto the LP of the 3D ANNNI model, the anisotropic 0 scaling of the two-point correlators could be con- 0 1 2 3 u1/2 firmed directly for the first time. The form of the scalingfunctionisinagreementwiththehypothe- sisofgeneralizedconformalinvariance. Fig. 2. Scaling funtion Φ(u) versus u1/2 = (√r⊥/rk)1/2 for κ=0.270 and T =3.7475. The different symbols cor- respondtodifferentvaluesofr⊥.Inset:comparisonofthe References full data set of 1.7 104 points for the scaling function × Φ(u)(graypoints)withtheanalyticalpredictionresulting fromreference [8](fullcurve). [1] R.M. Hornreich, M. Luban, and S. Shtrikman, Phys. Rev. Lett. 35,1678 (1975). and the scaling dimension x = (2+θ)/(2+γ/β) [2] W. Selke, Phys.Rep. 170,213 (1988). [8,9]. In this analysis, we use θ = 1/2, since the [3] B.Neubert,M.Pleimling,andR.Siems,Ferroelectrics smalldeviationsfromthisvalueobtainedinrecent 208-209,141 (1998). field-theoreticalcalculations[14,17]arenotyetdis- [4] W. Selke, Z.PhysikB29,133 (1978). tinguishable from the purely numerical errors in [5] K.KaskiandW.Selke,Phys.Rev.B31,3128(1985). our data. The data shownin Fig. 2 have been ob- tainedaftermorethanfivemillionclusterupdates [6] U. Wolff, Phys. Rev. Lett. 62,361 (1989). forasystemwith200×200×100spins.Theyper- [7] H.G.Evertzand W.v.d.Linden, Phys.Rev.Lett. 86, mit a nice visual test of the data collapse and es- 5164 (2001). tablishscaling,forthe firsttime alsoatthe LP. [8] M.Henkel, Phys.Rev. Lett. 78,1940 (1997). TheinsetofFig.2isadirectcomparisonofour [9] M. Pleimling and M. Henkel, Phys. Rev. Lett. 87, numericaldata with the theoreticalprediction re- 125702 (2001). sulting from a generalization of conformal invari- [10]W. Selke, in Phase Transitions and Critical ancetostronganisotropiccriticality[8].Appliedto Phenomena,Vol.15,ed.byC.DombandJ.L.Lebowitz theLPofthe3DANNNImodelandassumingθ = (Academic Press,New York,1992). 1/2, the scaling function Ω(v) = v−2x/θΦ(1/v) [11]M.Waasner and A.Hu¨ller,unpublished. shouldsatisfythe differentialequation [12]E. Luijten and H.W.J. Blo¨te, Int. J. Mod. Phys. C6, 359 (1995). d3Ω(v) dΩ(v) 2x α1 dv3 −v2 dv − θ vΩ(v)=0 (3) [13]J. Oitmaa, J. Phys.A: Math. Gen. 18,365 (1985). [14]H.W. Diehl and M. Shpot, Phys. Rev. B62, 12338 where α is a constant. It turns out [8,9] that the (2000). 1 functional form of Φ(u) only depends on a single [15]K.BinderandJ.-S.Wang,J.Stat.Phys.55,87(1989). universal parameter p which can be determined [16]M. Pleimling and W. Selke, Eur. Phys. J. B1, 385 from the Monte Carlo data by analyzing ratios of (1998). moments of Φ. As shown in [9], different moment [17]M. Shpot and H.W. Diehl, Nucl. Phys. B612, 340 ratios yield consistent values for this parameter. (2001). Themeanvalueofpobtainedinthiswayhasbeen 4

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.